ServiceRadar v1.4.70

Replacement for the incomplete 1.4.69 publication, including its
StarRocks flow analytics, device metrics, MTR and alerting changes.
Release builds now reject stale dependency metadata before publishing.

Whats New

1.4.70

• Release: refreshed Bazel dependency facts to match the existing Cargo
  lockfile. Application, native add-on and Wasm publishers now validate
  the committed dependency lockfile before publication and reject stale
  metadata instead of silently rewriting it during builds.
• Metrics: eligible short windows older than CNPG raw retention now read
  hourly aggregates instead of expired raw rows. The routing horizon
  defaults to 168 hours and can be set with
  SRQL_RAW_TELEMETRY_RETENTION_HOURS; flow routing is unchanged.
• Events: StarRocks event scalar values are truncated to their column
  limits at valid UTF-8 boundaries, with telemetry reporting truncation.
  This does not establish warehouse/CNPG count parity or enable cutover.
• Analytics: NetFlow reads now require StarRocks. Warehouse schema
  migration and retention support flows, scalar metrics, logs and events;
  dataset cutover remains explicit. Counter charts calculate rates from
  counter deltas, with fixes for tag-split series, newest-N limits, flow
  filters and dashboard readers following the configured backend.
• Topology: Dgraph deployment, schema provisioning and migration checks
  are available alongside AGE, with a Grafana dashboard for Dgraph.
• MTR: automated baseline, trigger and consensus processing is enabled
  by default. The new path analytics dashboard exposes packet-weighted
  loss and latency through SRQL, with time-bucketed trends and transit-ASN
  analysis. Built-in dashboard definitions preserve operator edits.
• Devices: bulk state editing, longer metric windows, and faster chart
  rendering. Inventory fixes reject reserved MAC identities, tighten
  duplicate-sweep policy and interface guards, and correct device
  pagination, vendor filters and event queries through identity aliases.
• Network changes: change-impact tracking records network changes and
  their affected devices.
• Alerts: stateful rules dispatch to the correct shard, handle structured
  grouping values without crashing, and serialize source UUIDs in both
  event details and diagnostics while preserving textual IDs.
• Dashboards: improved NetFlow loading, map arcs, Sankey edge selection
  and time-axis labels. The dashboard CLI resolves hoisted workspace
  dependencies and respects configured aliases during development.

Upgrade Notes

• Version 1.4.69 did not complete publication. Use 1.4.70 for the
  replacement release; its tag and any already-published 1.4.69 artifacts
  remain unchanged.
• When CNPG metric retention differs from seven days, set
  SRQL_RAW_TELEMETRY_RETENTION_HOURS to match and restart the process.
  Eligible old windows use hourly aggregate resolution; a window that
  crosses the raw retention horizon is served from aggregates throughout.
• NetFlow dashboards and in:flows no longer fall back to CNPG. Configure
  a working StarRocks warehouse and explicitly include flows in
  analytics.starrocks.cutoverDatasets before using flow reads. CNPG-only
  installations can continue reading metrics, logs and events.
• StarRocks remains optional in the base Helm values. Shared-nothing
  deployments use durable backend volumes; shared-data deployments need
  object storage. This release does not retire CNPG telemetry writes or
  automatically cut over every dataset. Warehouse migration can rebuild
  existing tables; allow capacity and time for that work.
• The Helm chart enables Dgraph by default and provisions TLS through
  cert-manager. Install cert-manager and provide persistent storage before
  upgrading with Dgraph enabled. Topology reads remain on AGE by default;
  the migration checksum must pass before deliberately switching reads.
  Set dgraph.enabled=false to retain an installation without an in-chart
  Dgraph cluster.
• Automated MTR processing can produce additional probes. To opt out, set
  MTR_AUTOMATION_ENABLED=false; explicitly configured baseline, trigger
  and consensus flags override that shared default and must also be
  disabled if set. MTR telemetry continues to use CNPG in this release.